k-Nearest Neighbours (KNN)¶
In [1]:
import warnings
warnings.filterwarnings('ignore')
In [2]:
from IPython.core.interactiveshell import InteractiveShell
InteractiveShell.ast_node_interactivity = 'all'
In [3]:
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
%matplotlib inline
In [4]:
df = pd.read_csv(r'C:\Users\..\KNNData.csv')
In [5]:
df
Out[5]:
In [6]:
df.describe()
Out[6]:
In [7]:
sns.pairplot(df)
Out[7]:
Data Pre-processing - Feature Scaling¶
1. Standard Scaler¶
$$ \frac{x_{i} - mean(x)}{stdev(x)} $$¶
In [8]:
from sklearn.preprocessing import StandardScaler
In [9]:
ss = StandardScaler()
In [10]:
scaled_features = ss.fit_transform(df.drop('TARGET CLASS', axis = 1))
In [11]:
np.round(pd.DataFrame(scaled_features).describe())
Out[11]:
2. Min-Max Scaler¶
$$\dfrac{x_{i} - min(x)}{max(x) - min(x)}$$¶
In [12]:
from sklearn.preprocessing import MinMaxScaler
In [13]:
mm = MinMaxScaler(feature_range=(0, 1)) # feature_range=(0, 1)
In [14]:
scaled_features_mm = mm.fit_transform(df.drop('TARGET CLASS', axis = 1))
In [15]:
np.round(pd.DataFrame(scaled_features_mm).describe())
Out[15]:
In [16]:
pd.DataFrame(scaled_features_mm)
Out[16]:
3. Robust Scaler¶
$$ \dfrac{x_{i} - Q_{1}(x)}{Q_{3}(x) - Q_{1}(x)} $$¶
In [17]:
from sklearn.preprocessing import RobustScaler
In [18]:
rr = RobustScaler()
In [19]:
scaled_features_rr = rr.fit_transform(df.drop('TARGET CLASS', axis = 1))
In [20]:
np.round(pd.DataFrame(scaled_features_rr).describe())
Out[20]:
In [21]:
pd.DataFrame(scaled_features_mm)
Out[21]:
In [22]:
sns.pairplot(pd.DataFrame(scaled_features_mm))
Out[22]:
k-NN Model Building¶
In [23]:
pd.DataFrame(scaled_features)
Out[23]:
In [25]:
df.columns[:-1]
Out[25]:
In [26]:
df_final = pd.DataFrame(scaled_features, columns=df.columns[:-1])
In [27]:
df_final
Out[27]:
In [29]:
df_final = pd.concat([df_final, df['TARGET CLASS']], axis = 1)
In [30]:
df_final
Out[30]:
In [31]:
from sklearn.model_selection import train_test_split
In [32]:
X = df_final.drop(['TARGET CLASS'], axis = 1)
Y = df_final['TARGET CLASS']
In [33]:
X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size = 0.3, random_state = 101)
Build 1NN (where k = 1)¶
In [34]:
from sklearn.neighbors import KNeighborsClassifier
In [36]:
knn1 = KNeighborsClassifier(n_neighbors=1)
In [37]:
knn1.fit(X_train, y_train)
Out[37]:
In [39]:
pred_knn1 = knn1.predict(X_test)
In [40]:
from sklearn import metrics
In [41]:
metrics.accuracy_score(y_test, pred_knn1)
Out[41]:
Applying GridSearchCV¶
In [42]:
from sklearn.model_selection import GridSearchCV
In [43]:
param_grid = {'n_neighbors' : range(3,17, 2), 'p' : [1,2]}
In [61]:
list( range(3,17, 2))
Out[61]:
In [48]:
knn = KNeighborsClassifier()
In [68]:
knn_gscv = GridSearchCV(knn, param_grid, cv=5,verbose=1)
In [69]:
import time
In [70]:
start_time = time.time()
knn_gscv.fit(X_train, y_train)
print(time.time() - start_time, 'Seconds')
Out[70]:
In [72]:
knn_gscv.best_params_
knn_gscv.best_score_
Out[72]:
Out[72]:
In [74]:
pd.DataFrame(knn_gscv.cv_results_)
Out[74]:
In [76]:
# Plot a graph to observe error rate
error_rate = []
for i in range(3,17, 2):
knn = KNeighborsClassifier(n_neighbors=i)
knn.fit(X_train, y_train)
pred = knn.predict(X_test)
error_rate.append(np.mean(pred != y_test))
Out[76]:
Out[76]:
Out[76]:
Out[76]:
Out[76]:
Out[76]:
Out[76]:
In [77]:
error_rate
Out[77]:
In [88]:
plt.figure(dpi = 125)
sns.lineplot(x =range(3,17, 2), y = error_rate,marker = 'o' , color = 'red');
In [ ]:
In [ ]:
In [ ]:
In [ ]:
In [ ]:
In [ ]:
In [ ]:
In [ ]:
In [ ]:
In [ ]:
In [ ]:
In [ ]: